Indonesian Journal of Statistics and Its Applications

Forecasting Foreign Tourists to West Sumatera Before and After COVID-19 Using ARIMA and Prophet and Its Impact on Foreign Exchange

Cindy Resha Alandra, Dony Permana, Dodi Vionanda, Dina Fitria — Tue, 30 Dec 2025 00:00:00 +0700

Foreign exchange earnings are very important for the improvement of the economy in Indonesia, where these foreign exchange earnings can be obtained through the tourism sector. One of the provinces in Indonesia that is a major tourist destination is West Sumatra. The number of foreign tourists coming to West Sumatra is influenced by various factors, one of which is the COVID-19 pandemic that resulted in a decrease in visitor numbers. The research was conducted to forecast the number of foreign tourists to West Sumatra using the ARIMA and Prophet methods, as well as to calculate the loss and foreign exchange earnings and the forecasting accuracy of both methods. The data for this study was taken from the BPS West Sumatra website regarding the number of foreign tourists to West Sumatra from 2015 to 2024. In this data, forecasting for the year 2020 will be done using the ARIMA method and forecasting for the year 2025 using the Prophet method. The data in this study tends to be stable before the pandemic, making the ARIMA method suitable. Meanwhile, after the pandemic, the data fluctuated, making the Prophet method suitable. From the results obtained, the best ARIMA model is ARIMA (1, 0, 1). The forecasting accuracy is 1.82% with an estimated foreign exchange loss of $52,095,688 for the year 2020. Meanwhile, using the Prophet method, the forecasting accuracy obtained is 12.13% with an estimated foreign exchange revenue of $208,546,812 for the year 2025.

Structural Equation Model Approach to National Health Insurance Participation in Disadvantaged Regions

Tue, 30 Dec 2025 00:00:00 +0700

Kartu Indonesia Sehat is Indonesia's national health insurance program, which is a right for everyone. However, the distribution of the program is not evenly distributed, especially in disadvantaged, frontier, and outermost regions. This research examines the factors influencing participation rates in Indonesia's national health insurance, the Healthy Indonesia Card program, in disadvantaged, frontier, and outermost regions of South Sulawesi. There are seven variables to estimate these factors, including education, employment, income, knowledge, motivation, socialization and trust. Based on descriptive statistics and a Structural Equation Model Partial Least Squares analysis using bootstrap parameters, there are three influencing factors: occupation, motivation, and trust, with a goodness of fit model of 32.3568%.

A Bayesian Approach for Wilcoxon Signed-Rank Test and Its Application to the Farmer’s Exchange Rate in Indonesia

Tridayanti Cahya Meiland, Suliadi Suliadi — Tue, 30 Dec 2025 00:00:00 +0700

Wilcoxon Signed-Rank Test (WSRT) is a rank-based statistical nonparametric method to test two paired samples. Researchers often use a frequentist approach in testing by utilizing test statistics or p-values. This approach has limitations in providing information about the rejection of the alternative or null hypothesis. These limitations have spurred interest in Bayesian-based testing, known as the Bayes Factor. The advantage of the Bayesian approach is it can measure how much the data support one hypothesis over another. However, there is a problem of using Bayesian approach in WSRT, since there is no distribution of the rank implies no likelihood can be formed from the data rank. Van Doorn proposed a Bayesian approach for this test by using a latent normal approach, by modeling the data rank being come from latent variables that are normally distributed. The objective of this study is to test whether there is difference between farmer exchange rate 2021 and 2022 in Indonesia. We used Wilcoxon Signed Rank Test with Bayesian approach as given by Van Doorn. The test employs Bayes factor to make conclusion, by transforming the rank of data using latent variable that assuming follow normal distribution. The analysis was conducted by constructing a posterior population of difference ( ) of 475,000 using the Gibbs Sampling algorithm. It is obtained the values of Bayes Factor is of 3076.07 and concluded that there is difference of farmer exchange rate in Indonesia between 2021 and 2022. This Bayes Factor indicates extreme evidence of a significant difference in the farmer exchange rate in Indonesia between 2021 and 2022.

Deep Learning–Based Semantic Segmentation for Evaluating Urban Environmental Quality and Walkability in Dongdaemun

Su Myat Thwin — Tue, 30 Dec 2025 00:00:00 +0700

This study evaluates environmental quality and urban walkability in the Dongdaemun district through geospatial semantic segmentation of street-view imagery. A DeepLab ResNet101 model, pre-trained on the ADE20K dataset and implemented using the GluonCV framework, was applied to Google Street View images collected at 40-meter intervals in four cardinal directions. Pixel-level segmentation was used to quantify key environmental features such as greenery, sky visibility, pavement, and road surfaces. Based on these visual attributes, composite indicators representing comfort, convenience, and safety were derived, leading to the calculation of an Integrated Visual Walkability index. The results reveal clear spatial variations in walkability across the study area, highlighting areas with favorable pedestrian environments and zones requiring improvement. Although the analysis is constrained by image quality and spatial coverage, the findings demonstrate the effectiveness of deep learning–based semantic segmentation for large-scale environmental assessment. This approach provides a scalable and data-driven framework to support evidence-based urban planning and sustainable city development.

Comparing Self-Paced Ensemble and RUSBoost for Imbalanced Poverty Classification in West Java

Nur Andi Setiabudi, Bagus Sartono, Utami Dyah Syafitri, Komang Budi Aryasa — Tue, 30 Dec 2025 00:00:00 +0700

Class imbalance remains a major challenge in classification modelling that frequently leads to biased predictive models. This study aimed to compare two ensemble techniques based on an undersampling approach, namely Self-Paced Ensemble and RUSBoost, for handling imbalanced classification in poverty identification in West Java. The results suggested that RUSBoost consistently outperformed Self-Paced Ensemble across the most critical metrics. It showed better balance in classification outcomes. When the objective is to maximize the identification of poor households, the default threshold in the RUSBoost model was prefered. On the other hand, if precision is prioritized due to limited resources, the Youden Index threshold offers a better alternative. Given the overall evaluation metrics, RUSBoost with the default threshold was suggested as the most reliable and well-balanced option among the compared models for classifying poor households in West Java under imbalanced data condition

OPEC Crude Oil Price Forecasting Using ARIMA with Ensemble Empirical Mode Decomposition

Tue, 30 Dec 2025 00:00:00 +0700

World crude oil prices fluctuate every day. One source of crude oil traded is oil from crude oil exporting countries that are members of the Organization of the Petroleum Exporting Countries (OPEC). In the total of 40% of world crude oil is produced by OPEC. This makes forecasting the price of crude oil OPEC’s policy very necessary in order to maintain world oil market stability. Fluctuating oil price data is made simpler and easier to interpret by applying the Ensemble Empirical Mode Decomposition (EEMD) method. The EEMD method decomposes the data into a number of Intrinsic Mode Functions (IMF) and residual of the IMF. In this study, the ARIMA forecasting model is compared using the original data and the decomposition results in the form of IMF components and IMF residuals. The comparison of the two methods is seen based on the overall and average MAPE value of the forecasting results in five time ranges. The EEMD-ARIMA method has an average MAPE value of 9.09% and standard deviation MAPE value of 7.39%. OPEC crude oil price forecast in January-August 2021 ranges from $42.22 to $60.6 per barrel. The final result of the analysis in this study shows that the ARIMA method with decomposition data (EEMD-ARIMA) is better than the ARIMA method using original data

Application of the Spatial Durbin Panel Model and Geographically Weighted Panel Regression on Poverty Data in West Java Province

Anis Sulistiyowati, Mohammad Masjkur, Budi Susetyo — Tue, 30 Dec 2025 00:00:00 +0700

Poverty is one of the priority issues in the Sustainable Development Goals. In 2024, West Java Province became the province with the second-highest number of people living in poverty in Indonesia. This study aims to identify the variables that significantly affect the percentage of people living in poverty in districts/cities of West Java Province from 2019 to 2023, using the spatial Durbin panel model and geographically weighted panel regression. The data used is secondary data on poverty indicators in West Java Province from 2019 to 2023, sourced from Statistics Indonesia of West Java. The spatial Durbin panel model developed in this study is a fixed-effects spatial Durbin panel model. The model shows that average years of schooling and expenditure per capita have significant effects. In addition, the spatial lags of the percentage of households living in appropriate housing, the percentage of the population covered by local health insurance, and average years of schooling also have significant effects. The geographically weighted panel regression model, estimated using a fixed effect panel regression with a Gaussian fixed kernel as the optimal weighting function, produces distinct models for each region. The average year of schooling is the dominant factor influencing the percentage of people living in poverty in districts/cities in West Java Province.

Cluster Level Time Series Forecasting on Indonesian Banking Stock Prices Using the Gated Recurrent Unit Method

Faisal Arkan, Budi Susetyo, Rahma Anisa — Tue, 30 Dec 2025 00:00:00 +0700

In recent years, there has been a significant increase in the number of Single Investor Identification registrations in the Indonesian capital market, as reported by the Financial Services Authority. Many investors favor stocks for their potential for high returns and liquidity. However, stock investments come with high risks due to their fluctuating prices, which are influenced by multiple factors. With 47 listed banking companies in the Indonesia Stock Exchange, clustering can help identify investor patterns. Forecasting stock prices is essential for anticipating future fluctuations. The large number of issuers and the tendency of stock prices to fluctuate increase the potential for outliers, requiring an appropriate clustering method. A study using the k-medoid method and dynamic time warping distance revealed 41 banking companies clustered into 5 clusters with a silhouette coefficient of 0.524. The Gated Recurrent Unit modeling, based on prototypes from the formed clusters, showed an excellent forecasting performance with root mean squared error and mean absolute percentage error ranging from 1-10%. The forecast for the next 8 weeks indicated varying price increases for each cluster. The first and third clusters are recommended for investors looking to maximize capital gains, due to their price increases and diverse cluster member characteristics. Additionally, investors should consider dividends provided by certain banking companies in their investment decision-making process.

Optimization of Fuzzy C-Means Clustering with Particle Swarm Optimization on Socioeconomic Indicators of ASEAN Countries

Tue, 30 Dec 2025 00:00:00 +0700

Grouping data based on similarity in characteristics is commonly applied in various exploratory analyses. The Fuzzy C-Means algorithm offers flexibility through the degree of membership of data points in each cluster, but it is vulnerable to poor cluster center initialization, which increases the risk of getting trapped in local optima. To enhance the performance of Fuzzy C-Means, this study integrates the Particle Swarm Optimization method for determining cluster centers. The evaluation is conducted by comparing Fuzzy C-Means and Fuzzy C-Means-Particle Swarm Optimization across several cluster counts using three internal validation metrics, namely the silhouette coefficient, partition coefficient, and Xie-Beni Index. The results show that Fuzzy C-Means-Particle Swarm Optimization consistently yields higher silhouette coefficient and partition coefficient values, along with lower Xie-Beni Index values, compared to standard Fuzzy C-Means. This indicates that the integration of Particle Swarm Optimization can improve clustering quality in terms of cluster compactness and separation. This hybrid approach demonstrates significant potential in complex data clustering scenarios.